Publications

Measuring Progress in Fine-grained Vision-and-Language Understanding
Emanuele Bugliarello
Laurent Sartran
Aishwarya Agrawal
Lisa Anne Hendricks
Aida Nematzadeh
While pretraining on large-scale image–text data from the Web has facilitated rapid progress on many vision-and-language (V&L) tasks, rece… (voir plus)nt work has demonstrated that pretrained models lack “fine-grained” understanding, such as the ability to recognise relationships, verbs, and numbers in images. This has resulted in an increased interest in the community to either develop new benchmarks or models for such capabilities. To better understand and quantify progress in this direction, we investigate four competitive V&L models on four fine-grained benchmarks. Through our analysis, we find that X-VLM (Zeng et al., 2022) consistently outperforms other baselines, and that modelling innovations can impact performance more than scaling Web data, which even degrades performance sometimes. Through a deeper investigation of X-VLM, we highlight the importance of both novel losses and rich data sources for learning fine-grained skills. Finally, we inspect training dynamics, and discover that for some tasks, performance peaks early in training or significantly fluctuates, never converging.
2023-06-30
Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (publié)
doi.org
arxiv.org
Optimal Transport for Unsupervised Hallucination Detection in Neural Machine Translation
Nuno M. Guerreiro
Pierre Colombo
Pablo Piantanida
André Martins
Neural machine translation (NMT) has become the de-facto standard in real-world machine translation applications. However, NMT models can un… (voir plus)predictably produce severely pathological translations, known as hallucinations, that seriously undermine user trust. It becomes thus crucial to implement effective preventive strategies to guarantee their proper functioning. In this paper, we address the problem of hallucination detection in NMT by following a simple intuition: as hallucinations are detached from the source content, they exhibit encoder-decoder attention patterns that are statistically different from those of good quality translations. We frame this problem with an optimal transport formulation and propose a fully unsupervised, plug-in detector that can be used with any attention-based NMT model. Experimental results show that our detector not only outperforms all previous model-based detectors, but is also competitive with detectors that employ external models trained on millions of samples for related tasks such as quality estimation and cross-lingual sentence similarity.
2023-06-30
Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (publié)
doi.org
arxiv.org
Reinforcement Learning-Based Adaptive Feature Boosting for Smart Grid Intrusion Detection
Chengming Hu
Jun Yan
Xue Liu
Intrusion detection systems (IDSs) are crucial in the security monitoring for the smart grid with increasing machine-to-machine communicatio… (voir plus)ns and cyber threats thereafter. However, the multi-sourced, correlated, and heterogeneous smart grid data pose significant challenges to the accurate attack detection by IDSs. To improve the attack detection, this paper proposes Reinforcement Learning-based Adaptive Feature Boosting, which aims to leverage a series of AutoEncoders (AEs) to capture critical features from the multi-sourced smart grid data for the classification of normal, fault, and attack events. Multiple AEs are utilized to extract representative features from different feature sets that are automatically generated through a weighted feature sampling process; each AE-extracted feature set is then applied to build a Random Forest (RF) base classifier. In the feature sampling process, Deep Deterministic Policy Gradient (DDPG) is introduced to dynamically determine the feature sampling probability based on the classification accuracy. The critical features that improve the classification accuracy are assigned larger sampling probabilities and increasingly participate in the training of next AE. The presence of critical features is increased in the event classification over the multi-sourced smart grid data. Considering potential different alarms among base classifiers, an ensemble classifier is further built to distinguish normal, fault, and attack events. Our proposed approach is evaluated on the two realistic datasets collected from Hardware-In-the-Loop (HIL) and WUSTIL-IIOT-2021 security testbeds, respectively. The evaluation on the HIL security dataset shows that our proposed approach achieves the classification accuracy with 97.28%, an effective 5.5% increase over the vanilla Adaptive Feature Boosting. Moreover, the proposed approach not only accurately and stably selects critical features on the WUSTIL-IIOT-2021 dataset based on the significant difference of feature sampling probabilities between critical and uncritical features, i.e., the probabilities greater than 0.08 and less than 0.01, but also outperforms the other best-performing approaches with the increasing Matthew Correlation Coefficient (MCC) of 8.03%.
2023-06-30
IEEE Transactions on Smart Grid (publié)
doi.org
SkillQG: Learning to Generate Question for Reading Comprehension Assessment
Xiaoqiang Wang
Bang Liu
Siliang Tang
Lingfei Wu
2023-06-30
Findings of the Association for Computational Linguistics: ACL 2023 (publié)
doi.org
arxiv.org
The KITMUS Test: Evaluating Knowledge Integration from Multiple Sources
Akshatha Arodi
Martin Pömsl
Kaheer Suleman
Adam Trischler
A.R. Olteanu
Jackie CK Cheung
Many state-of-the-art natural language understanding (NLU) models are based on pretrained neural language models. These models often make in… (voir plus)ferences using information from multiple sources. An important class of such inferences are those that require both background knowledge, presumably contained in a model’s pretrained parameters, and instance-specific information that is supplied at inference time. However, the integration and reasoning abilities of NLU models in the presence of multiple knowledge sources have been largely understudied. In this work, we propose a test suite of coreference resolution subtasks that require reasoning over multiple facts. These subtasks differ in terms of which knowledge sources contain the relevant facts. We also introduce subtasks where knowledge is present only at inference time using fictional knowledge. We evaluate state-of-the-art coreference resolution models on our dataset. Our results indicate that several models struggle to reason on-the-fly over knowledge observed both at pretrain time and at inference time. However, with task-specific training, a subset of models demonstrates the ability to integrate certain knowledge types from multiple sources. Still, even the best performing models seem to have difficulties with reliably integrating knowledge presented only at inference time.
2023-06-30
Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (publié)
doi.org
arxiv.org
EuclidNets: An Alternative Operation for Efficient Inference of Deep Learning Models
Xinlin Li
Mariana Parazeres
Adam Oberman
Alireza Ghaffari
Masoud Asgharian
Vahid Nia
2023-06-29
SN Computer Science (publié)
doi.org
arxiv.org
Reference panel-guided super-resolution inference of Hi-C data
Yanlin Zhang
Mathieu Blanchette
Abstract Motivation Accurately assessing contacts between DNA fragments inside the nucleus with Hi-C experiment is crucial for understanding… (voir plus) the role of 3D genome organization in gene regulation. This challenging task is due in part to the high sequencing depth of Hi-C libraries required to support high-resolution analyses. Most existing Hi-C data are collected with limited sequencing coverage, leading to poor chromatin interaction frequency estimation. Current computational approaches to enhance Hi-C signals focus on the analysis of individual Hi-C datasets of interest, without taking advantage of the facts that (i) several hundred Hi-C contact maps are publicly available and (ii) the vast majority of local spatial organizations are conserved across multiple cell types. Results Here, we present RefHiC-SR, an attention-based deep learning framework that uses a reference panel of Hi-C datasets to facilitate the enhancement of Hi-C data resolution of a given study sample. We compare RefHiC-SR against tools that do not use reference samples and find that RefHiC-SR outperforms other programs across different cell types, and sequencing depths. It also enables high-accuracy mapping of structures such as loops and topologically associating domains. Availability and implementation https://github.com/BlanchetteLab/RefHiC.
2023-06-29
Bioinformatics (publié)
doi.org
Should We Feed the Trolls? Using Marketer-Generated Content to Explain Average Toxicity and Product Usage
Marcelo Vinhal Nepomuceno
Hooman Rahemi
Tolga Cenesizoglu
Laurent Charlin
2023-06-28
Journal of Interactive Marketing (publié)
doi.org
Cortico-Cerebellar neurodynamics during social interaction in Autism Spectrum Disorders
Fleur Gaudfernau
Aline Lefebvre
Denis-Alexander Engemann
Amandine Pedoux
Anna Bánki
Florence Baillin
Benjamin Landman
Frederique Amsellem
Anna Maruani
Thomas Bourgeron
Richard Delorme
Guillaume Dumas
2023-06-27
NeuroImage : Clinical (publié)
doi.org
Pixelated Reconstruction of Foreground Density and Background Surface Brightness in Gravitational Lensing Systems using Recurrent Inference Machines
Alexandre Adam
Laurence Perreault-Levasseur
Yashar Hezaveh
MAX WELLING
Modeling strong gravitational lenses in order to quantify the distortions in the images of background sources and to reconstruct the mass de… (voir plus)nsity in the foreground lenses has been a difficult computational challenge. As the quality of gravitational lens images increases, the task of fully exploiting the information they contain becomes computationally and algorithmically more difficult. In this work, we use a neural network based on the Recurrent Inference Machine (RIM) to simultaneously reconstruct an undistorted image of the background source and the lens mass density distribution as pixelated maps. The method iteratively reconstructs the model parameters (the image of the source and a pixelated density map) by learning the process of optimizing the likelihood given the data using the physical model (a ray-tracing simulation), regularized by a prior implicitly learned by the neural network through its training data. When compared to more traditional parametric models, the proposed method is significantly more expressive and can reconstruct complex mass distributions, which we demonstrate by using realistic lensing galaxies taken from the IllustrisTNG cosmological hydrodynamic simulation.
2023-06-26
The Astrophysical Journal (publié)
doi.org
arxiv.org
Adaptive Discrete Communication Bottlenecks with Dynamic Vector Quantization for Heterogeneous Representational Coarseness
Dianbo Liu
Alex Lamb
Xu Ji
Pascal Junior Tikeng Notsawo
Michael Mozer
Yoshua Bengio
Kenji Kawaguchi
Vector Quantization (VQ) is a method for discretizing latent representations and has become a major part of the deep learning toolkit. It ha… (voir plus)s been theoretically and empirically shown that discretization of representations leads to improved generalization, including in reinforcement learning where discretization can be used to bottleneck multi-agent communication to promote agent specialization and robustness. The discretization tightness of most VQ-based methods is defined by the number of discrete codes in the representation vector and the codebook size, which are fixed as hyperparameters. In this work, we propose learning to dynamically select discretization tightness conditioned on inputs, based on the hypothesis that data naturally contains variations in complexity that call for different levels of representational coarseness which is observed in many heterogeneous data sets. We show that dynamically varying tightness in communication bottlenecks can improve model performance on visual reasoning and reinforcement learning tasks with heterogeneity in representations.
2023-06-25
AAAI Conference on Artificial Intelligence (publié)
doi.org
arxiv.org
Bugs in the Data: How ImageNet Misrepresents Biodiversity
Alexandra Luccioni
David Rolnick
ImageNet-1k is a dataset often used for benchmarking machine learning (ML) models and evaluating tasks such as image recognition and object … (voir plus)detection. Wild animals make up 27% of ImageNet-1k but, unlike classes representing people and objects, these data have not been closely scrutinized. In the current paper, we analyze the 13,450 images from 269 classes that represent wild animals in the ImageNet-1k validation set, with the participation of expert ecologists. We find that many of the classes are ill-defined or overlapping, and that 12% of the images are incorrectly labeled, with some classes having >90% of images incorrect. We also find that both the wildlife-related labels and images included in ImageNet-1k present significant geographical and cultural biases, as well as ambiguities such as artificial animals, multiple species in the same image, or the presence of humans. Our findings highlight serious issues with the extensive use of this dataset for evaluating ML systems, the use of such algorithms in wildlife-related tasks, and more broadly the ways in which ML datasets are commonly created and curated.
2023-06-25
Proceedings of the AAAI Conference on Artificial Intelligence (publié)
doi.org
arxiv.org